An ensemble technique for estimating vehicle speed and gear position from acoustic data

This paper presents a machine learning system that is capable of predicting the speed and gear position of a moving vehicle from the sound it makes. While audio classification is widely used in other research areas such as music information retrieval and bioacoustics, its application to vehicle sounds is rare. Therefore, we investigate predicting the state of a vehicle using audio features in a classification task. We improve the classification results using correlation matrices, calculated from signals correlating with the audio. In an experiment, the sound of a moving vehicle is classified into discretized speed intervals and gear positions. The experiment shows that the system is capable of predicting the vehicle speed and gear position with near-perfect accuracy over 99%. These results show that this system could be a valuable addition to vehicle anomaly detection and safety systems.